Ikiru

As AI tools get better at producing a first draft, a working piece of code, or a plausible answer, the value of producing those things at all quietly declines. What becomes more valuable is something less visible: the ability to tell whether what was produced is actually correct, appropriate, and worth using.

Generation is getting cheap; evaluation is not

A model can produce a plausible-sounding paragraph, function, or analysis in seconds. Whether that output is actually right, in a specific context, still requires someone who understands the subject well enough to spot the subtle error, the missing edge case, or the confident-sounding claim that does not quite hold up. That evaluative skill was always valuable. It is becoming the bottleneck, now that production is no longer one.

People with strong foundational knowledge in a field are positioned to use AI tools as genuine accelerators, since they can quickly spot when something is off and correct it.

This creates an uncomfortable dependency: evaluating AI output well requires roughly the same depth of understanding it would take to produce the output without help in the first place. Someone who never built that underlying understanding has no real way to catch a plausible-sounding mistake, because the mistake is specifically designed, by nature of how these tools work, to sound reasonable.

If you tracked every hour spent helping a colleague move, reviewing a friend's portfolio, or introducing two people who might get along professionally, it would look like a poor use of time on any given day. None of it pays immediately. Most of it is never repaid by the same person it was given to.

Why the accounting looks wrong at short range

Reciprocity rarely works one-to-one. The person you help this month is not usually the person who helps you next month. Instead, favors tend to circulate through a wider group over years, in forms that do not resemble the original favor at all — a recommendation from someone you barely know, an opportunity that traces back three or four introductions. Judged transaction by transaction, generosity looks unrewarding. Judged over a decade inside a stable group, it tends to be one of the better returns available.

Most discussion about AI in education focuses on cheating and generated essays. A quieter and arguably larger shift is happening underneath that debate: students now have the option to skip the uncomfortable early stage of learning where nothing makes sense yet, and that stage was doing more work than it appeared to.

Confusion was never just an obstacle

Before instant, clear answers were available, a confused student had to sit with a problem for a while — rereading, guessing, testing an idea and watching it fail. That sitting-with-confusion was uncomfortable and also produced something real: a stronger, more durable model of the material, built through the friction of getting it wrong first. Skipping straight to a clear explanation removes the discomfort and, often, some of that durability along with it.

Most independent creatives track exactly one financial number: how much came in this month. That number feels informative and is actually one of the least useful ones available, because it says nothing about whether the business behind it is getting stronger or quietly running on fumes.

Effective hourly rate, not project rate

A project priced at a flat fee can hide a wildly unprofitable hourly rate if scope creeps, which it usually does. Dividing total payment by actual hours spent, including revisions and calls, tends to be uncomfortable the first time it is calculated honestly. It is also the single most useful number for deciding which kinds of projects to keep taking and which to quietly stop accepting, regardless of how impressive they look on a portfolio.